Why You Trust Patterns That Don't Exist

A woman in her mid-thirties walked into a sixth specialist’s office with joint pain, recurring fevers, and a rash that came and went. Five previous doctors had told her it was lupus. Each had prescribed steroids, watched her improve temporarily, and sent her home. Eighteen months after her first visit, a rheumatologist at Johns Hopkins finally ordered a test for adult-onset Still’s disease—a rare autoimmune condition that mimics lupus but requires different treatment. The test came back positive. Why did six specialists miss it? Not because they were careless. Not because they lacked intelligence. They missed it because their brains were doing exactly what evolution designed them to do: pattern-match. Every doctor had seen dozens of lupus cases. Lupus is common. Still’s disease is not. When a patient presents with joint pain, fever, and rash, the statistical probability tilts heavily toward lupus. The pattern fits. The diagnosis feels right. And feeling right is exactly the problem. The same cognitive machinery that makes expert clinicians fast and accurate for common cases also makes them dangerously blind to low-probability events. This is not a failure of knowledge. It is a failure built into the very structure of how we learn from experience. In the eighteenth century, the Scottish philosopher David Hume pointed out a logical hole in human reasoning: we have no rational basis for assuming the future will resemble the past. Every time we see the sun rise, we expect it to rise again tomorrow—but nothing in logic guarantees it. We simply feel that it will. Hume called this the problem of induction, and it is not an abstract puzzle. It is the operating system of every expert’s mind. Pattern recognition is fast, unconscious, and energy-efficient. It works beautifully for situations where the past is a reliable guide to the future. If you touch a hot stove, you learn not to touch it again. That’s induction in its most useful form. But when the situation is rare, when the evidence is ambiguous, when the cost of being wrong is high, the same mechanism becomes a trap. The Sarah case—let’s call her that—illustrates the trap with painful clarity. Each specialist saw symptoms that fit a common script. They ordered the standard lupus panel. It came back borderline or negative, but that didn’t break the pattern—lupus can be seronegative. They prescribed treatment. The patient improved partially. That partial improvement confirmed the diagnosis. The pattern held. The test for Still’s disease sat in a lab manual, never ordered, because it never entered the mental script. This is not a story about bad doctors. It is a story about how expertise itself creates blind spots. The more cases a clinician has seen, the more deeply the pattern is etched. Their pattern-recognition system becomes exquisitely tuned to the frequencies they encounter most often. That tuning is what makes them fast and effective for the 95% of cases that are common. But it also makes them slower to recognize the 5% that are not. The same dynamic plays out in investing, where past returns become a proxy for future performance. It plays out in relationships, where a partner’s past behavior becomes the lens through which all future actions are interpreted. It plays out in any domain where humans make high-stakes decisions under uncertainty. We trust the pattern because the pattern has worked before. But the pattern has no logical authority. It has only history. Awareness of this bias is not enough. Knowing about Hume’s problem of induction does not automatically slow down your pattern-matching system. The brain’s fast track is always on. The challenge is to build deliberate friction into the process when the stakes are high. One technique that works, in medicine and elsewhere, is to force yourself to generate alternative explanations before settling on the first fit. The rheumatologist who tested for Still’s disease did not simply trust the pattern. She asked: What if this is not lupus? What else could cause this constellation? That question is simple, but it is not natural. It requires slowing down, resisting the satisfaction of closure, and holding uncertainty open longer than feels comfortable. Systemic constraints make this harder. Short appointment times, productivity pressure, and the sheer volume of cases push clinicians toward fast pattern-matching. The same is true for investors facing quarterly earnings calls or managers making hiring decisions. The system rewards speed and penalizes deliberation. But the cost of a wrong pattern match can be eighteen months of misdiagnosis, or a portfolio wiped out by an event that “never happened before.” The takeaway is not to abandon pattern recognition—that would be like asking a fish to stop swimming. The takeaway is to recognize when the pattern is most likely to fool you: when the case is rare, when the evidence is ambiguous, when the outcome of being wrong is catastrophic. In those moments, deliberately pause. Ask what you are not seeing. The pattern you trust most may be the one that leads you furthest from the truth.